[llvm] r251291 - Loop Vectorizer - skipping "bitcast" before GEP

[Demikhovsky, Elena via llvm-commits]

This test checks remark messages. It expects to see “the cost-model indicates that vectorization is not beneficial”.
But after skipping  ‘bitcast’ the vectorization becomes beneficial and the expected message is not generated any more.

This is the original code that wasn’t vectorized due to the ‘bitcast’:

  %arrayidx = getelementptr inbounds float, float* %in, i64 %indvars.iv, !dbg !9
  %0 = bitcast float* %arrayidx to i32*, !dbg !9
  %1 = load i32, i32* %0, align 4, !dbg !9, !tbaa !11

I just took another loop in order to get the expected remark.


-           Elena

From: Eric Christopher [mailto:echristo at gmail.com]
Sent: Wednesday, October 28, 2015 03:25
To: Demikhovsky, Elena; llvm-commits at lists.llvm.org
Subject: Re: [llvm] r251291 - Loop Vectorizer - skipping "bitcast" before GEP


 ; First loop.
 ;  #pragma clang loop interleave(disable) unroll(disable)
 ;  for(int i = 0; i < n; i++) {
-;    out[i] = in[i];
+;    out[i] = *in[i];
 ;  }

 ; Second loop.
 ;  #pragma clang loop unroll(disable)
 ;  for(int i = 0; i < n; i++) {
-;    out[i] = in[i];
+;    out[i] = *in[i];
 ;  }

As a start in looking at this (since it was reverted, but before you apply again), why did you regenerate the testcase with different beginning code? It doesn't make sense.

-eric
---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151028/053a2a8c/attachment.html>